[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[TIGGE #YSF-211129]: TCP tuning for TIGGE tuning

This archive contains answers to questions sent to Unidata support through mid-2025. Note that the archive is no longer being updated. We provide the archive for reference; many of the answers presented here remain technically correct, even if somewhat outdated. For the most up-to-date information on the use of NSF Unidata software and data services, please consult the Software Documentation first.

  • Subject: [TIGGE #YSF-211129]: TCP tuning for TIGGE tuning
  • Date: Tue, 13 Feb 2007 11:22:53 -0700
Hi YangXin,

Thanks for the update regarding the new hardware you're using.  Please let us 
know the output from "uname -a"
Network performance tuning can cause unexpected problems, so we'll try changing 
only the variables that will 
give you the largest improvements.  Try adding the following to 
/etc/sysctl.conf and running "sysctl -p"
after the changes are made;

net.core.rmem_max = 1000000
net.core.wmem_max = 1000000
net.ipv4.tcp_rmem = 4096 524288 2000000
net.ipv4.tcp_wmem = 4096 65536 2000000

Afterwards, please stop and re-start the LDM.  That should make a noticable 
improvement to performance.  
If for some reason it seems to make performance worse, change the vaules in 
/etc/sysctl.conf back to the 
default and re-run "sysctl -p"  Let me know if that works or if it doesn't

mike

> Hi, Steve,
> 
> The current platform for TIGGE Data Exchange in CMA is a IBM x3850 equipped 
> with 2 2-core processors and 4GB memory, 
> running RHEL4 update 4. As you suggested, TCP Tuning should have significant 
> impact to the efficiency of data transfer,
> I'd like to do some test to change the important kernel parameters, see if 
> the performance can be improved and less 
> data will be lost.
> 
> I'm sorry that I don't have a deep understanding on network performance 
> tuning, so I have to check some books and 
> materials to find out what is the critical factor to the networking system. 
> I've found serveral parameters that 
> probably have impact on TCP performance, but I'm not very sure what value is 
> best for LDM to work for our platform. 
> These parameters with the default value in our platform are as follows:
> 
> net.ipv4.tcp_tw_reuse = 0
> net.ipv4.tcp_tw_recycle = 0
> 
> net.ipv4.tcp_fin_timeout = 60
> 
> net.ipv4.tcp_max_syn_backlog = 1024
> 
> net.ipv4.tcp_rmem = 4096        87380   174760
> net.core.rmem_default = 135168
> net.core.rmem_max = 135168
> 
> net.ipv4.tcp_wmem = 4096        16384   131072
> net.core.wmem_default = 135168
> net.core.wmem_max = 135168
> 
> 
> I believe there must be some other parameters that are important to TCP 
> Tuning. I only enabled 
> first two parameter by setting them as follows:
> net.ipv4.tcp_tw_reuse = 1
> net.ipv4.tcp_tw_recycle = 1
> 
> and then restarted LDM.
> 
> Is it appropriate, do I need to set any other parameters?
> 
> Thank you.
> 
> YangXin,
> Feb 13.

Ticket Details
===================
Ticket ID: YSF-211129
Department: Support IDD TIGGE
Priority: High
Status: Closed